87 
rate, though twice as large here, was still acceptable at 8 percent, 
and a small increase for crop/pasture. The correspondences of the 
forest, crop/pasture, and water categories were in excellent agreement 
0> 95 percent), as before, with the (P' + S') adjustment. Impres 
sively, LDR, which formerly was associated with a 97 percent error 
rate, was reduced to less than a 5 percent disagreement by the 
(P* + S') adjustment. And the 30 percent error rate for the CII cate 
gory was reduced to an acceptable 8 percent. The transitional category 
was also much improved, from 33.5 to 15.2 percent error. 
Comparison of the MAGI statistics for P and (P* + S') revealed that 
reassignments made as P was adjusted to (P’ + S') for each cover type 
were: +2.9 percent for forest, -12.7 percent for crop/pasture, -50.0 
percent for water, -15.9 percent for transitional, -2.9 percent for 
CII, +4.6 percent for MDR, and +13.1 percent for LDR. The most notable 
differences between these figures and those computed for the whole area 
were associated with the downward adjustments to the transitional and 
CII categories of 8 and 3 times former adjustments respectively. 
Optimal MAGI System/Landsat Data Correspondence 
The results from Tables I and II are summarized in Table III where the 
computation options for the MAGI System data were ranked in order of 
determination complexity. From this analysis, neither the P nor P' 
options, whether for the whole area or the more cover-specific subarea, 
provide the required minimum level of accuracy 0> 90 percent). The 
adjusted data (P' + S') were required in general, and specific land use 
labels were required to achieve high correspondence for CII and LDR and 
near acceptable correspondence for the transitional category. 
Spatial Degradation of Landsat Data 
The results of resampling the Landsat data used in Table II (the 85 
percent subarea) to produce a spatially degraded product (L R ) of the 
same grid cell size as the MAGI 4.6 acre data base were: forest, 
36,790 acres (58.4 percent); crop/pasture, 7,944 acres (12.6 percent); 
water, 63 acres (0.1 percent); transitional, 602 acres (1.0 percent); 
CII, 4,623 acres (7.3 percent); MDR, 9,278 (14.7 percent); and LDR, 
3,684 acres (5.8 percent). These results were compared with the 
earlier full resolution Landsat data (L ), and with the MAGI data (M). 
F 
The resampling of the 1.54 acre Landsat data to 4.6 acre cells created 
a significant degradation in correspondence between the original Land 
sat data and the MAGI data, as shown in Tables IV and V. CII and LDR 
were the only categories to maintain a reasonable correspondence 
between the original and resampled Landsat data sets (Lj,:L^). Good 
correspondence occurred in the MDR and LDR categories for the L^:M com 
parison, but only for the P level of MAGI data, a consequence of 
removing information content in the degradation process. A similar 
pattern exists for the L :L and L :M comparisons: error rates in the 
F R R 
15-20 percent range for forest, crop/pasture, and water and in excess 
of 40 percent for transitional. This similarity is not surprising 
given the good correspondence of L :M. 
r 
DISCUSSION 
The results indicate that for good correspondence of the Landsat and 
MAGI System data for the seven land cover categories examined here, the 
(P' + S') information level from the MAGI System is preferred for esti 
mating all cover types. The computation of (P' + S') is especially